One of the best practices we know from great engineers is the back-of-the-envelope calculation to estimate costs and resources. In Machine Learning Engineering, we all would benefit from such a “back-of-the-envelope calculation” skill to design an ML system. We need to confirm - as cheaply as possible - that our future ML project is worthwhile and will solve a business problem, and that the costs and resources will be feasible. In this talk, I present a collaborative design toolkit for ML projects that supports identifying ML use cases and performing rough prototyping by using three canvases: Machine Learning Canvas, Data Landscape Canvas, and MLOps Stack Canvas.